On the same day Anthropic filed for its IPO, OpenAI released GPT-6.1 Terra and...

On October 26 local time, on the same day that Anthropic submitted its IPO registration filing to the U.S. Securities and Exchange Commission (SEC), OpenAI released its new model GPT-6.1 Terra, cutting the standard API price by 50% compared with the previous generation.

OpenAI said that GPT-6.1 Terra focuses on visual understanding and computer operation, and can handle tasks such as filling out web forms, editing videos, and reading technical drawings. In the Agents' Last Exam professional task evaluation, the model improved by 15 percentage points over GPT-5.6 Terra at medium reasoning effort, while the cost per single task dropped to about one-third; in the Terminal-Bench 4.0 coding evaluation, its score at the highest reasoning effort was 50.3%, higher than the previous generation's 21.5%.

Pricing is the focus of this release. GPT-6.1 Terra's standard API input price is $1 per million tokens, output price is $5, and cached input is $0.05. Compared with GPT-6 Astra, its input and output token unit prices are both one-tenth; compared with GPT-5.6 Terra, the standard price is down 50%.

OpenAI said that Terra has approached flagship-model level in some task evaluations, but the most complex coding tasks are still better suited to GPT-6.1 Sol or GPT-6 Astra.

This is not a simple price-cutting promotion, nor is it the first time OpenAI has pursued a price-cutting marketing strategy. Early on, OpenAI's price cuts mostly came alongside model upgrades. Since 2025, due to the impact of open-source models, OpenAI's price-cutting strategy has become more aggressive and direct, with the goal clearly aimed at capturing call volume and ecosystem share. For example, the output price of the reasoning model o3 was cut by as much as 80%, and the input price of GPT-5.6 Luna dropped from $1 per million tokens to $0.20.

Behind the price cuts is the company's expectation of larger-scale call volume. OpenAI CEO Sam Altman attaches considerable importance to the prospect of Terra 6.1 "ushering in the era of computer use."

For computer operation and agent applications to truly enter enterprise workflows, models must be able to reliably complete multi-step tasks while pushing the inference cost of a single task down to a sustainable level. For developers, model capability being close does not mean the actual cost is the same. Lower token prices and task costs help expand the scope of use of automation applications and drive growth in call volume.

OpenAI's choice this time to strengthen computer operation capabilities while simultaneously cutting prices reflects its strategy of trying to win high-frequency enterprise tasks at lower cost.

This strategy is also related to the funding pressure OpenAI faces. According to market reports, OpenAI's cumulative cash burn from 2026 to 2030 could reach $278 billion, mainly related to computing power and infrastructure investment. Although the company has already made preparations for a potential listing, Altman said it will not go public in 2026. For OpenAI, which still needs to continue investing in frontier research and computing power, how to convert model capabilities into stable revenue and improve the commercial return per unit of computing power is a question that the capital markets continue to focus on.

As OpenAI's largest competitor, Anthropic's prospectus presents a similar picture from the financial side: revenue is growing rapidly and quarterly operations are improving, but the long-term cost pressure brought by computing power investment remains.

The document shows that Anthropic's revenue in the first half of 2026 reached $16.286 billion, far higher than $1.163 billion in the same period last year; second-quarter operating profit was $974 million, with an operating margin of about 8.4%; after excluding items such as stock-based compensation, adjusted operating profit was $2.235 billion, with an adjusted operating margin of about 19.3%.

Although the net loss calculated under US generally accepted accounting principles in the first half of the year reached $80.748 billion, a considerable portion of it was not cash expenditure for the period.

However, long-term computing power procurement and leasing commitments still constitute fixed cost pressure. Even with rapid revenue growth, the company will still need to continue investing substantial funds in the future to support model training and customer demand. In other words, Anthropic likewise needs to find a balance between capability and revenue growth on one hand and computing power costs and commercialization returns on the other, and use this to prove operational improvement to the capital markets.